San Francisco-based startup Niobia is addressing one of the most costly problems in battery and chip manufacturing: understanding the causes of defects that appear on production lines. The company uses AI software to detect defects, then allows engineers to ask a chatbot why they occurred and analyze the associated data within minutes instead of spending days on manual investigation.
Companies using the technology deploy AI agents on production lines to monitor defects, while human engineers remain part of the process of identifying the likely cause and understanding the failure conditions. These agents operate on standalone GPUs installed directly on manufacturing equipment, functioning in a way similar to a sensor, which allows them to detect extremely small defects, including scratches or holes no larger than the scale of a microscopic pore.
From Detecting a Defect to Explaining Its Cause
Niobia’s function is not limited to reporting that a fault exists. The platform brings the relevant data together in one place, then allows engineers to ask follow-up questions through a chatbot to understand what happened and explore the potential causes in greater depth. The company says this approach is intended to reduce the time required to restart production lines after a problem emerges during research and development or manufacturing process improvements.
Speed is particularly important in the battery and semiconductor industries, where every error can lead to costly downtime and delays in manufacturing schedules, with losses reaching millions of dollars in some cases. The complexity of materials and technology development processes also makes identifying the cause of a fault more difficult than simply detecting a nonconforming product.
On-Premises AI to Protect Manufacturing Data
Niobia relies on open-source models that it retrains using data and research related to batteries and chips. According to co-founder Gaurav Jha, the company chose this approach because customers do not want to send their sensitive industrial data to a public cloud, even when there are commitments to protect it.
The company allows customers to retain their own encryption keys, meaning Niobia cannot decrypt their data. This reflects the fact that the platform’s requirements concern not only the accuracy of defect detection, but also where the data is processed and who has the ability to access it, particularly in industries characterized by intellectual property sensitivity and geopolitical competition.
Initial Focus and Ambition Toward More Autonomous Factories
The product currently focuses on three manufacturing processes: film coating; electroplating, which uses an electric current to cover surfaces with metals; and fluidized-bed reaction, a chemical process in which solid particles are suspended in a gas or liquid. Jha said the company later wants to increase the autonomy of its technology, ultimately reaching factories capable of operating themselves to a greater degree.
Why does this matter? The practical value of Niobia’s offering is not replacing engineers, but reducing the time between detecting a defect and understanding its cause. However, the source provides no figures on the platform’s performance, the number of customers, or detection accuracy, so its actual commercial and industrial impact still requires verification through independent operational results. The ambition of reaching a self-operating factory also remains a future goal, rather than a fully announced capability today.